In　order　to　improve　the　performance　of　time-series　classification,　we　introduce　a　new　approach　of　time　series　classification.　The　first　basic　idea　of　the　approach　is　to　use　entropy　impurity　measure　to　discretize　and　symbolize　time　series,　which　discretize　the　original　time　series　into　disjoint　intervals　using　entropy　impurity　measure　and　then　transform　the　time　series　into　symbolic　representations.　The　second　idea　of　the　approach　is　to　combine　symbolic　representation　of　time　series　and　k　nearest　neighbor　to　classify　time　series.　The　proposed　approach　is　compared　with　a　number　of　known　pattern　classifiers　by　benchmarking　with　the　use　of　artificial　and　real-world　data　sets.　The　experimental　results　show　it　can　reduce　the　error　rates　of　time　series　classification,　so　it　is　highly　competitive　with　previous　approaches.